Pith. sign in

REVIEW 4 major objections 4 minor 25 references

Classification of Linear Observed Systems on Multi-Frame Groups via Automorphisms

T0 review · 4 major / 4 minor · reviewed 2026-08-11 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The new multi-frame group classifies every linear observed system into six process and six observation forms.

desk verdict The multi-frame construction is real and useful, but Theorem 2 is false—MFG is isomorphic to a direct product of TFGs, so the automorphism group includes block permutations and the claimed classification is incomplete. read the letter →

arxiv 2412.14673 v1 pith:TCOLMTDL submitted 2024-12-19 eess.SY cs.SY

Changwu Liu , Yuan Shen This is my paper · ORCID
classification eess.SYcs.SY MSC 93C1093B27
keywords multi-framegrouplinearobservedsystemsaffinedynamicsalgebraicobservationsautomorphisminvariantextendedKalmanfilternavigationtwo-frame
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper constructs a multi-frame group (MFG) by chaining two-frame groups through semi-direct products, so that the pose states of several coupled frames live in one Lie group. It then tries to classify every linear observed system on this group: every group-affine process equation must take one of the forms (19)--(24), and every algebraic observation one of the forms (30)--(35). The classification is carried out by a direct calculation through the automorphism group of MFG rather than by case-by-case geometric construction. If correct, any multi-frame navigation problem whose equations match these forms can inherit the group structure and with it the consistency and stability guarantees of invariant filters. As an illustration, the paper applies the construction to depth-camera inertial odometry with online camera--IMU extrinsic calibration.

What carries the argument

The load-bearing object is the multi-frame group $\mathrm{MFG}(d,n,m,s,t)$, a type-II semi-direct product of $s+1$ left and $t+1$ right copies of the two-frame group $\mathrm{TFG}(d,n,m)$, glued by the inner-automorphism action. The argument runs through the automorphism group: Theorem 2 asserts $\mathrm{Aut}(\mathrm{MFG})$ is exactly the set of conjugations $\psi_S(\chi)=S\chi S^{-1}$ with $S=\mathrm{diag}(S_1,\ldots,S_{s+t+1})$ and each $S_i$ an upper-triangular block matrix with diagonal blocks $\Omega_i\in\mathrm{SO}(d)$ and $A_i\in\mathrm{GL}(n+m,\mathbb{R})$, hence $S_i\in\mathrm{SIM}_{n+m}(d)$. This single structure does double duty: it fixes the form of every group-affine flow $\Phi_t(\chi_0)=S\chi_0 S^{-1}\Phi_t(\mathrm{id})$, and, via the lemma that $\phi(g)\triangleright p$ is a new action whenever $\phi$ is an automorphism, it fixes the possible algebraic observations. Writing $S$ and $W=S^{-1}\Phi_t(\mathrm{id})$ in block form and equating coefficients produces the classified ODEs and observation equations.

What would settle it

Compute the automorphism group of the smallest non-trivial case $\mathrm{MFG}(2,1,1,1,1)$ explicitly; if it contains any automorphism not of the form $\chi \mapsto S\chi S^{-1}$ with $S=\mathrm{diag}(S_1,S_2,S_3)$ and each $S_i$ an upper-triangular block matrix of the stated SIM form, then Theorem 2 collapses and the classification misses systems.

Watch

Extended reading notes

Core claim

The central claim is that the multi-frame group $\mathrm{MFG}(d,n,m,s,t)$ supports a complete classification of linear observed systems: for any choice of parameters, the group-affine process dynamics reduce to the explicit ODEs (19)--(24), and the left- and right-invariant algebraic observations reduce to (30)--(32) and (33)--(35), with all coefficients freely chosen functions of the control input. The classification is achieved by proving that every automorphism of $\mathrm{MFG}$ is a block-diagonal conjugation by $S=\mathrm{diag}(S_1,\ldots,S_{s+t+1})$ with each $S_i\in\mathrm{SIM}_{n+m}(d)$, so every group-affine flow has the form $\Phi_t(\chi_0)=S\chi_0 S^{-1}\Phi_t(\mathrm{id})$. Equating blocks along this flow yields the announced forms, and the same conjugation structure, applied through a lemma that turns automorphisms into new group actions, yields the observation forms. The paper presents the classification as exhaustive: a system not matching these forms cannot be made linear observed under the MFG state structure.

Load-bearing premise

The classification is complete only if the multi-frame group has no symmetries beyond the block-diagonal conjugations listed in Theorem 2; the paper gives no proof beyond 'Direct calculation verifies the theorem,' so the whole list of possible systems rests on that unstated check.

Editorial extensions

If this is right

  • A multi-frame navigation system whose process and observation equations match one of the listed forms is automatically a linear observed system on MFG, so an invariant extended Kalman filter can be applied with the state-independent error Jacobians that give guaranteed local stability.
  • The classification is exhaustive: if a system's equations do not match forms (19)--(24) and (30)--(35), no group structure of MFG type can make it linear observed under the paper's automorphism characterization.
  • Setting $j=0$ in the classified dynamics recovers the natural vector dynamics of the two-frame group from [17], so the multi-frame classification contains the earlier two-frame classification as a special case.
  • All coefficient matrices and vectors in the classified forms are arbitrary functions of the control input, which gives the designer the full freedom to fit sensor models while preserving the linear-observed property.
  • In the worked example, the depth-camera observation matches the right-action form (35), so the MFG construction applies and yields a filter whose extrinsics error decays faster in transient response than the multiplicative EKF or an imperfect invariant EKF.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • One consequence the author leaves implicit is that the same automorphism-based calculation should classify linear observed systems on any group built by repeated semi-direct products from a base group, not just TFG; testing it on SE(3)-based building blocks or on groups with bias states is a natural next step.
  • The completeness of the list is only as strong as the automorphism theorem, and a reader who wants to rely on the classification would need the explicit calculation that the paper compresses into 'Direct calculation verifies the theorem.'
  • Because the simulation treats the filters as deterministic observers, a stochastic Monte Carlo consistency study (for example, normalized estimation error squared) would test whether the better transient behavior of the MFG-IEKF persists under realistic noise.
  • The classification doubles as a no-go test: a proposed multi-frame sensor model that does not fit the listed forms cannot be made linear observed under an MFG structure, so practitioners should look for a different symmetry group instead of forcing the fit.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 4 minor

Summary. The paper proposes a multi-frame group (MFG) constructed via a semi-direct product of two-frame groups TFG(d,n,m), gives a block-diagonal matrix embedding, and states that the automorphism group of MFG consists of block-diagonal conjugations by S = diag(S_1,...,S_{s+t+1}) with each S_i in SIM_{n+m}(d) (Theorem 2). It then classifies all group-affine process ODEs (Theorem 3, eqs. (19)-(24)) and all algebraic observations (Theorem 4, eqs. (30)-(35)) on MFG using this automorphism structure, and demonstrates an application to depth-camera inertial odometry with online extrinsic calibration. The central claim is that this provides a systematic and complete classification of linear observed systems on multi-frame groups.

Significance. If correct, the classification would substantially simplify the modeling of multi-sensor navigation problems as linear observed systems, enabling invariant-filter guarantees without case-by-case constructions. The paper's application section and simulation give a concrete demonstration of the potential practical value. However, the entire classification rests on Theorem 2, which asserts a characterization of Aut(MFG) that is false. Because the automorphism group is larger than the block-diagonal conjugation family, the derived ODE and observation forms do not exhaust all linear observed systems on MFG. The theoretical contribution is therefore not valid as stated; the application may still be salvageable, but the completeness claim is unsupported.

major comments (4)
  1. [Section III, Theorem 2] The claimed characterization of Aut(MFG) is false. By Theorem 1, the change of coordinates B_0 = T_0, B_j = lT_j B_{j-1} (1 ≤ j ≤ s), B_{s+j} = B_{s+j-1} rT_j (1 ≤ j ≤ t) is a group isomorphism from MFG(d,n,m,s,t) to the direct product TFG(d,n,m)^{s+t+1} with componentwise multiplication. Consequently, for any permutation π ∈ S_{s+t+1}, the map that permutes the diagonal blocks of χ = diag(B_0,...,B_{s+t}) is an automorphism of MFG. For s=1,t=0, the map σ : diag(T_0, lT_1 T_0) ↦ diag(lT_1 T_0, T_0) is an automorphism that cannot be realized as SχS^{-1} with S = diag(S_1,S_2) and S_i ∈ SIM_{n+m}(d), because conjugation by a block-diagonal S acts independently on each diagonal block. Thus the automorphism group is strictly larger than the family asserted in Theorem 2, and the one-line proof 'Direct calculation verifies the theorem' is insufficient and incorrect.
  2. [Section IV.A, eq. (5) and Theorem 3] Because Theorem 2 is false, the reduction of all group-affine dynamics to eq. (5), Φ_t(χ0) = Sχ0S^{-1}Φ_t(id), is incomplete. For the factor-permutation automorphism σ with s=1,t=0, the flow Φ_t(B_0,B_1) = (B_1(0), B_0(0)) · (P_0(t), P_1(t)) yields an ODE in which ˙B_0 = B_1(t) P_1(t)^{-1} ˙P_0(t), so the derivative of B_0 depends on B_1(t). In contrast, every equation in (19)-(24) gives ˙B_j as a function of B_j and (for composite blocks) terms involving earlier blocks through Ad, but not a dependence of B_0 on B_1. The classification therefore omits legitimate group-affine dynamics obtained by composing with factor-permutation automorphisms, and the abstract's claim to classify all possible forms is not substantiated.
  3. [Section III, Definition 3 and following text] The text asserts that the choice φ(T) = ψ_T (the inner automorphism map) 'supports the coupling of multiple frames' and that the construction 'covers all natural extensions.' However, Theorem 1 shows that MFG is isomorphic to the direct product TFG^{s+t+1}; the semi-direct product is trivialized by the change of variables to the diagonal-block representation. The physical chain structure is preserved in the interpretation of the blocks, but the group-theoretic construction is not a genuinely coupled semi-direct product. This does not invalidate the application, but the conceptual claim that the inner-automorphism twist introduces coupling between frames is misleading and should be corrected.
  4. [Section IV, proofs of Theorems 3 and 4] The derivations of the classification theorems are summarized as 'equating blocks by brute calculation' (Theorem 3) and 'simplifying the equations... we have proved the theorem' (Theorem 4). For a classification result that constitutes the main contribution, this level of detail is inadequate, especially because the omitted automorphism-group analysis is precisely where the error in Theorem 2 occurs. A rigorous proof or a detailed appendix is needed, and the current presentation does not allow the reader to verify the completeness claim.
minor comments (4)
  1. [Section IV.A, after eq. (7)] There is a typo: 'simn+m(d) of SIMn+d(d)' should read 'sim_{n+m}(d) of SIM_{n+m}(d)'.
  2. [Section IV.A, eqs. (6) and (15)] The relation between S_i and W_i (with lower-right blocks inverse to each other) is used later, but the introduction of lL0 = -rL0 in (15) is abrupt; a brief derivation of this constraint from (6) would improve readability.
  3. [Section V, Fig. 2] The simulation plots only the camera-IMU extrinsic errors. Reporting the full state errors (attitude, position, velocity) would strengthen the demonstration, although the current figure is acceptable for a letter.
  4. [Section IV.B, Theorem 4] The statement of Theorem 4 is long and notationally dense; an explicit example of the new summation notations (e.g., for s=1,t=1) would help the reader verify the forms (30)-(35).

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the classification is derived from an explicit automorphism parameterization with free coefficient functions, not from fitted data or load-bearing self-citation.

full rationale

The paper's derivation chain is self-contained rather than circular. The multi-frame group is defined explicitly (Definition 3), its matrix embedding is stated and verified (Theorem 1), and the automorphism group is asserted as block-diagonal conjugation (Theorem 2). The group-affine dynamics are then written as Phi_t(chi0) = S chi0 S^{-1} Phi_t(id) in equation (5), and the process ODE forms (19)-(24) are obtained by differentiating this parameterization. The observation forms (30)-(35) follow from applying automorphisms to the natural matrix action via Lemma 1. In all of these steps, the matrices S, W, and the constants in Theorem 4 are free functions or free constants; they are not fitted to data, and no subset of data is used to predict a closely related quantity. The simulation uses tunable filter gains and preset trajectories, not fitted parameters, so there is no fitted-input-called-prediction issue. The only self-reference is to the authors' prior work [2] on linear observed systems on manifolds with connection, but that citation is background context and not the load-bearing justification for the MFG classification, which rests on the paper's own Theorem 2 calculation and the external framework [1]. The proof of Theorem 2 is terse ('Direct calculation verifies the theorem'), which is an omitted-support and potential correctness concern, especially if outer automorphisms such as block permutations exist; however, that is a mathematical correctness or completeness risk, not a circular reduction of the conclusion to its inputs. No equation is defined in terms of the result it is supposed to derive, and no cited uniqueness theorem from the authors is invoked to forbid alternatives. Therefore the circularity score is 0.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

The theoretical classification contains no fitted constants; all matrix-valued coefficients are arbitrary functions of control inputs. The simulation uses manually tuned covariance gains, but those do not support the central claim. The multi-frame group is a mathematical construction, not a postulated physical entity; no new particles, forces, or dimensions are introduced.

assumptions (4)
  • domain assumption Linear observed systems have group-affine flows of the form Φ_t(χ0)=ψ_{u_t}(χ0)Φ_t(id) with ψ in Aut(G).
    Taken from prior theory [1]; the paper uses this as the definition of group affine dynamics in Section IV-A.
  • ad hoc to paper Aut(MFG) is exactly the block-diagonal conjugation group in Theorem 2.
    Stated as Theorem 2 with a one-line proof; the entire classification depends on it.
  • domain assumption All algebraic observations are generated by the natural matrix action of MFG on V=R^{m+n+d}, possibly twisted by automorphisms via Lemma 1.
    The paper considers only the left/right matrix action and automorphism twists; it does not prove all group actions are of this form (Section IV-B).
  • ad hoc to paper The MFG construction itself, via the canonical homomorphism φ(T)=ψ_T (inner automorphism), is the natural 'covering' of multi-frame extensions.
    Definition 3; the authors choose inner automorphisms to create coupling between frames.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Classification of Linear Observed Systems on Multi-Frame Groups via Automorphisms." pith.science (2026). https://pith.science/paper/TCOLMTDL

@misc{pith2026241214673,
  author       = {Pith},
  title        = {Pith review of: Classification of Linear Observed Systems on Multi-Frame Groups via Automorphisms},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TCOLMTDL}},
  note         = {Machine review of arXiv:2412.14673}
}
read the original abstract

Many navigation problems can be formulated as observer design on linear observed systems with a two-frame group structure, on which an invariant filter can be implemented with guaranteed consistency and stability. It's still unclear how this could be generalized to simultaneous estimation of the poses of multiple frames and the general forms of the linear observed systems involving multiple frames remain unknown. In this letter, we propose a multi-frame group structure by semi-direct product using the two-frame group as building blocks, covering all natural extensions. More importantly, we give a systematic direct calculation to classify all possible forms of linear observed systems including process ODEs and algebraic observations on such multi-frame group through its automorphism structure, in comparison to the existing classification on two-frame groups relying on ingenious construction. Depth-camera inertial odometry with online extrinsics calibration is provided as an application.

Figures

Figures reproduced from arXiv: 2412.14673 by the authors.

Figure 1
Figure 1. Transformation chain structure of MFG. Each node represents a [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Plot of error w.r.t time of MEKF, imperfect IEKF (IEKF) and [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

25 extracted references · 18 canonical work pages

  1. [1]

    Linear observed systems on groups,

    A. Barrau and S. Bonnabel, “Linear observed systems on groups,” Systems & Control Letters , vol. 129, pp. 36–42, July 2019

    work page 2019

  2. [2]

    On the existence of linear observed systems on manifolds with connection,

    C. Liu and Y . Shen, “On the existence of linear observed systems on manifolds with connection,” IEEE Control Systems Letters, vol. 8, pp. 2607–2612, 2024

    work page 2024

  3. [3]

    The invariant extended kalman filter as a stable observer,

    A. Barrau and S. Bonnabel, “The invariant extended kalman filter as a stable observer,” IEEE Transactions on Automatic Control , vol. 62, no. 4, pp. 1797–1812, 2017

    2017

  4. [4]

    Invariant kalman filtering,

    A. Barrau and S. Bonnabel, “Invariant kalman filtering,” Annual Review of Control, Robotics, and Autonomous Systems , vol. 1, pp. 237–257, 2018

    2018

  5. [5]

    Equivariant filter (eqf),

    P. van Goor, T. Hamel, and R. Mahony, “Equivariant filter (eqf),” IEEE Transactions on Automatic Control , vol. 68, no. 6, pp. 3501–3512, 2023

    work page 2023

  6. [6]

    Equivariant filter (eqf): A general filter design for systems on homogeneous spaces,

    P. van Goor, T. Hamel, and R. Mahony, “Equivariant filter (eqf): A general filter design for systems on homogeneous spaces,” in 2020 59th IEEE Conference on Decision and Control (CDC) , 2020, pp. 5401–5408

    work page 2020

  7. [7]

    Equivariant systems theory and observer design,

    R. Mahony, T. Hamel, and J. Trumpf, “Equivariant systems theory and observer design,” 2020. [Online]. Available: https: //arxiv.org/abs/2006.08276

    arXiv 2020

  8. [8]

    Equivariant filter design for kinematic systems on lie groups,

    R. Mahony and J. Trumpf, “Equivariant filter design for kinematic systems on lie groups,” IFAC-PapersOnLine, vol. 54, no. 9, pp. 253– 260, 2021, 24th International Symposium on Mathematical Theory of Networks and Systems MTNS 2020

    work page 2021

Show all 25 references
  1. [9]

    Eqvio: An equivariant filter for visual- inertial odometry,

    P. van Goor and R. Mahony, “Eqvio: An equivariant filter for visual- inertial odometry,” IEEE Transactions on Robotics, vol. 39, no. 5, pp. 3567–3585, 2023

    2023

  2. [10]

    Msceqf: A multi state constraint equivariant filter for vision-aided inertial navigation,

    A. Fornasier, P. van Goor, E. Allak, R. Mahony, and S. Weiss, “Msceqf: A multi state constraint equivariant filter for vision-aided inertial navigation,” IEEE Robotics and Automation Letters , vol. 9, no. 1, pp. 731–738, 2024

    2024

  3. [11]

    Decoupled right invariant error states for consistent visual-inertial navigation,

    Y . Yang, C. Chen, W. Lee, and G. Huang, “Decoupled right invariant error states for consistent visual-inertial navigation,” IEEE Robotics and Automation Letters , vol. 7, no. 2, pp. 1627–1634, 2022

    2022

  4. [12]

    Ingvio: A consistent invariant filter for fast and high-accuracy gnss-visual-inertial odometry,

    C. Liu, C. Jiang, and H. Wang, “Ingvio: A consistent invariant filter for fast and high-accuracy gnss-visual-inertial odometry,” IEEE Robotics and Automation Letters , vol. 8, no. 3, pp. 1850–1857, 2023

    2023

  5. [13]

    Exploiting symmetries to design ekfs with consistency properties for navigation and slam,

    M. Brossard, A. Barrau, and S. Bonnabel, “Exploiting symmetries to design ekfs with consistency properties for navigation and slam,”IEEE Sensors Journal, vol. 19, no. 4, pp. 1572–1579, 2019

    2019

  6. [14]

    Autonomous error and constructive ob- server design for group affine systems,

    P. van Goor and R. Mahony, “Autonomous error and constructive ob- server design for group affine systems,” in 2021 60th IEEE Conference on Decision and Control (CDC) , 2021, pp. 4730–4737

    2021

  7. [15]

    Constructive equivariant observer design for inertial navigation,

    P. van Goor, T. Hamel, and R. Mahony, “Constructive equivariant observer design for inertial navigation,” IFAC-PapersOnLine, vol. 56, no. 2, pp. 2494–2499, 2023, 22nd IFAC World Congress

    2023

  8. [16]

    Synchronous observer design for inertial navigation systems with almost-global convergence,

    P. van Goor, T. Hamel, and R. Mahony, “Synchronous observer design for inertial navigation systems with almost-global convergence,” 2023. [Online]. Available: https://arxiv.org/abs/2311.02234

    2023 arXiv

  9. [17]

    The geometry of navigation problems,

    A. Barrau and S. Bonnabel, “The geometry of navigation problems,” IEEE Transactions on Automatic Control, vol. 68, no. 2, pp. 689–704, 2023

    2023

  10. [18]

    Optimal active sens- ing control for two-frame systems,

    J. Benhamou, S. Bonnabel, and C. Chapdelaine, “Optimal active sens- ing control for two-frame systems,” in 2023 62nd IEEE Conference on Decision and Control (CDC) , 2023, pp. 5967–5974

    2023

  11. [19]

    Invariant filtering for wheeled vehicle localization with unknown wheel radius and unknown gnss lever arm,

    P. Chauchat, S. Bonnabel, and A. Barrau, “Invariant filtering for wheeled vehicle localization with unknown wheel radius and unknown gnss lever arm,” 2024. [Online]. Available: https: //arxiv.org/abs/2409.07050

    2024 arXiv

  12. [20]

    Invariant kalman filtering with noise-free pseudo-measurements,

    S. Goffin, S. Bonnabel, O. Brls, and P. Sacr, “Invariant kalman filtering with noise-free pseudo-measurements,” in2023 62nd IEEE Conference on Decision and Control (CDC) , 2023, pp. 8665–8671

    2023

  13. [21]

    J. M. Lee, Introduction to Smooth Manifolds . New York: Springer, 2003

    2003

  14. [22]

    B. C. Hall, Lie Groups, Lie Algebras, and Representations: An Elementary Introduction. Switzerland: Springer, 2015

    2015

  15. [23]

    Equivariant symmetries for aided inertial navigation,

    A. Fornasier, “Equivariant symmetries for aided inertial navigation,”

  16. [24]

    Overcoming bias: Equivariant filter design for biased attitude estima- tion with online calibration,

    A. Fornasier, Y . Ng, C. Brommer, C. Bhm, R. Mahony, and S. Weiss, “Overcoming bias: Equivariant filter design for biased attitude estima- tion with online calibration,” IEEE Robotics and Automation Letters , vol. 7, no. 4, pp. 12 118–12 125, 2022

    2022

  17. [2024]

    Available: https://arxiv.org/abs/2407.14297

    [Online]. Available: https://arxiv.org/abs/2407.14297

    arXiv

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.